Reads your code, silently judges your life choices, and roasts every bad line.
It doesn't ask "are you ready for feedback?" It asks "are you emotionally stable enough to handle this?"
Linters catch syntax and miss judgment. Review tools that do catch judgment produce polite reports nobody opens. The analysis is rarely what fails — the delivery is.
So CodeRoast reports through the editor's own Diagnostics API, the same channel a real linter uses. Findings appear squiggled under the offending line, in the Problems panel, where you are already looking — not in a separate report you will close.
The humour is the delivery mechanism. I act on considerably more of this feedback than I ever did on a clean report, which was the entire hypothesis.
Logic flaws · naming crimes · empty catch blocks · missing error handling · style · things
that would make your senior dev sigh audibly.
if (data == null || data == undefined) {Congratulations on checking for null twice and undefined zero times.
==already did this for you. You wrote extra characters to achieve nothing.
Inline diagnostics, plus a roast-summary.md written to .code-roast/database/ with severity
counts, recurring habits and suggested improvements — so the output is auditable afterwards
rather than just momentary.
Bring your own Gemini key. No code is routed through any server of mine.
Settings → Extensions → CodeRoast, or in settings.json:
{ "codeRoast.geminiApiKey": "YOUR_API_KEY_HERE" }No key, no roast. Gemini demands tribute.
Built by Uday Parmar